CIMGEN: Controlled Image Manipulation by Finetuning Pretrained Generative Models on Limited Data

Content creation and image editing can benefit from flexible user controls. A common intermediate representation for conditional image generation is a semantic map, that has information of objects present in the image. When compared to raw RGB pixels, the modification of semantic map is much easier. One can take a semantic map and easily modify the map to selectively insert, remove, or replace objects in the map. The method proposed in this paper takes in the modified semantic map and alter the original image in accordance to the modified map. The method leverages traditional pre-trained image-to-image translation GANs, such as CycleGAN or Pix2Pix GAN, that are fine-tuned on a limited dataset of reference images associated with the semantic maps. We discuss the qualitative and quantitative performance of our technique to illustrate its capacity and possible applications in the fields of image forgery and image editing. We also demonstrate the effectiveness of the proposed image forgery technique in thwarting the numerous deep learning-based image forensic techniques, highlighting the urgent need to develop robust and generalizable image forensic tools in the fight against the spread of fake media.

AdaTreeFormer: Few Shot Domain Adaptation for Tree Counting from a Single High-Resolution Image

The process of estimating and counting tree density using only a single aerial or satellite image is a difficult task in the fields of photogrammetry and remote sensing. However, it plays a crucial role in the management of forests. The huge variety of trees in varied topography severely hinders tree counting models to perform well. The purpose of this paper is to propose a framework that is learnt from the source domain with sufficient labeled trees and is adapted to the target domain with only a limited number of labeled trees. Our method, termed as AdaTreeFormer, contains one shared encoder with a hierarchical feature extraction scheme to extract robust features from the source and target domains. It also consists of three subnets: two for extracting self-domain attention maps from source and target domains respectively and one for extracting cross-domain attention maps. For the latter, an attention-to-adapt mechanism is introduced to distill relevant information from different domains while generating tree density maps; a hierarchical cross-domain feature alignment scheme is proposed that progressively aligns the features from the source and target domains. We also adopt adversarial learning into the framework to further reduce the gap between source and target domains. Our AdaTreeFormer is evaluated on six designed domain adaptation tasks using three tree counting datasets, ie Jiangsu, Yosemite, and London; and outperforms the state of the art methods significantly.

Improving 2D-3D Dense Correspondences with Diffusion Models for 6D Object Pose Estimation

Estimating 2D-3D correspondences between RGB images and 3D space is a fundamental problem in 6D object pose estimation. Recent pose estimators use dense correspondence maps and Point-to-Point algorithms to estimate object poses. The accuracy of pose estimation depends heavily on the quality of the dense correspondence maps and their ability to withstand occlusion, clutter, and challenging material properties. Currently, dense correspondence maps are estimated using image-to-image translation models based on GANs, Autoencoders, or direct regression models. However, recent advancements in image-to-image translation have led to diffusion models being the superior choice when evaluated on benchmarking datasets. In this study, we compare image-to-image translation networks based on GANs and diffusion models for the downstream task of 6D object pose estimation. Our results demonstrate that the diffusion-based image-to-image translation model outperforms the GAN, revealing potential for further improvements in 6D object pose estimation models.

Knowledge-driven deep learning for fast MR imaging: undersampled MR image reconstruction from supervised to un-supervised learning

Deep learning (DL) has emerged as a leading approach in accelerating MR imaging. It employs deep neural networks to extract knowledge from available datasets and then applies the trained networks to reconstruct accurate images from limited measurements. Unlike natural image restoration problems, MR imaging involves physics-based imaging processes, unique data properties, and diverse imaging tasks. This domain knowledge needs to be integrated with data-driven approaches. Our review will introduce the significant challenges faced by such knowledge-driven DL approaches in the context of fast MR imaging along with several notable solutions, which include learning neural networks and addressing different imaging application scenarios. The traits and trends of these techniques have also been given which have shifted from supervised learning to semi-supervised learning, and finally, to unsupervised learning methods. In addition, MR vendors' choices of DL reconstruction have been provided along with some discussions on open questions and future directions, which are critical for the reliable imaging systems.

Quantified Task Misalignment to Inform PEFT: An Exploration of Domain Generalization and Catastrophic Forgetting in CLIP

Foundations models are presented as generalists that often perform well over a myriad of tasks. Fine-tuning these models, even on limited data, provides an additional boost in task-specific performance but often at the cost of their wider generalization, an effect termed catastrophic forgetting. In this paper, we analyze the relation between task difficulty in the CLIP model and the performance of several simple parameter-efficient fine-tuning methods through the lens of domain generalization and catastrophic forgetting. We provide evidence that the silhouette score of the zero-shot image and text embeddings is a better measure of task difficulty than the average cosine similarity of correct image/label embeddings, and discuss observable relationships between task difficulty, fine-tuning method, domain generalization, and catastrophic forgetting. Additionally, the averaged results across tasks and performance measures demonstrate that a simplified method that trains only a subset of attention weights, which we call A-CLIP, yields a balance between domain generalization and catastrophic forgetting.

Switch EMA: A Free Lunch for Better Flatness and Sharpness

Exponential Moving Average (EMA) is a widely used weight averaging (WA) regularization to learn flat optima for better generalizations without extra cost in deep neural network (DNN) optimization. Despite achieving better flatness, existing WA methods might fall into worse final performances or require extra test-time computations. This work unveils the full potential of EMA with a single line of modification, i.e., switching the EMA parameters to the original model after each epoch, dubbed as Switch EMA (SEMA). From both theoretical and empirical aspects, we demonstrate that SEMA can help DNNs to reach generalization optima that better trade-off between flatness and sharpness. To verify the effectiveness of SEMA, we conduct comparison experiments with discriminative, generative, and regression tasks on vision and language datasets, including image classification, self-supervised learning, object detection and segmentation, image generation, video prediction, attribute regression, and language modeling. Comprehensive results with popular optimizers and networks show that SEMA is a free lunch for DNN training by improving performances and boosting convergence speeds.

An advanced data fabric architecture leveraging homomorphic encryption and federated learning

Data fabric is an automated and AI-driven data fusion approach to accomplish data management unification without moving data to a centralized location for solving complex data problems. In a Federated learning architecture, the global model is trained based on the learned parameters of several local models that eliminate the necessity of moving data to a centralized repository for machine learning. This paper introduces a secure approach for medical image analysis using federated learning and partially homomorphic encryption within a distributed data fabric architecture. With this method, multiple parties can collaborate in training a machine-learning model without exchanging raw data but using the learned or fused features. The approach complies with laws and regulations such as HIPAA and GDPR, ensuring the privacy and security of the data. The study demonstrates the method's effectiveness through a case study on pituitary tumor classification, achieving a significant level of accuracy. However, the primary focus of the study is on the development and evaluation of federated learning and partially homomorphic encryption as tools for secure medical image analysis. The results highlight the potential of these techniques to be applied to other privacy-sensitive domains and contribute to the growing body of research on secure and privacy-preserving machine learning.

GaMeS: Mesh-Based Adapting and Modification of Gaussian Splatting

Recently, a range of neural network-based methods for image rendering have been introduced. One such widely-researched neural radiance field (NeRF) relies on a neural network to represent 3D scenes, allowing for realistic view synthesis from a small number of 2D images. However, most NeRF models are constrained by long training and inference times. In comparison, Gaussian Splatting (GS) is a novel, state-of-the-art technique for rendering points in a 3D scene by approximating their contribution to image pixels through Gaussian distributions, warranting fast training and swift, real-time rendering. A drawback of GS is the absence of a well-defined approach for its conditioning due to the necessity to condition several hundred thousand Gaussian components. To solve this, we introduce the Gaussian Mesh Splatting (GaMeS) model, which allows modification of Gaussian components in a similar way as meshes. We parameterize each Gaussian component by the vertices of the mesh face. Furthermore, our model needs mesh initialization on input or estimated mesh during training. We also define Gaussian splats solely based on their location on the mesh, allowing for automatic adjustments in position, scale, and rotation during animation. As a result, we obtain a real-time rendering of editable GS.

InternLM-XComposer2: Mastering Free-form Text-Image Composition and Comprehension in Vision-Language Large Model

We introduce InternLM-XComposer2, a cutting-edge vision-language model excelling in free-form text-image composition and comprehension. This model goes beyond conventional vision-language understanding, adeptly crafting interleaved text-image content from diverse inputs like outlines, detailed textual specifications, and reference images, enabling highly customizable content creation. InternLM-XComposer2 proposes a Partial LoRA (PLoRA) approach that applies additional LoRA parameters exclusively to image tokens to preserve the integrity of pre-trained language knowledge, striking a balance between precise vision understanding and text composition with literary talent. Experimental results demonstrate the superiority of InternLM-XComposer2 based on InternLM2-7B in producing high-quality long-text multi-modal content and its exceptional vision-language understanding performance across various benchmarks, where it not only significantly outperforms existing multimodal models but also matches or even surpasses GPT-4V and Gemini Pro in certain assessments. This highlights its remarkable proficiency in the realm of multimodal understanding. The InternLM-XComposer2 model series with 7B parameters are publicly available at https://github.com/InternLM/InternLM-XComposer.

Understanding Neural Network Systems for Image Analysis using Vector Spaces and Inverse Maps

There is strong interest in developing mathematical methods that can be used to understand complex neural networks used in image analysis. In this paper, we introduce techniques from Linear Algebra to model neural network layers as maps between signal spaces. First, we demonstrate how signal spaces can be used to visualize weight spaces and convolutional layer kernels. We also demonstrate how residual vector spaces can be used to further visualize information lost at each layer. Second, we introduce the concept of invertible networks and an algorithm for computing input images that yield specific outputs. We demonstrate our approach on two invertible networks and ResNet18.